| Functionality | netCDF | SD | AIO | HDF5 | Comments |
|---|---|---|---|---|---|
| Open existing file for read/write | ncopen | SDstart | AIO_open | H5Fopen | |
| Creates new file for read/write. | nccreate | H5Fcreate | SD API handles this with SDopen | ||
| Close file | ncclose | SDend | AIO_close | H5Fclose | |
| Redefine parameters | ncredef | Unneccessary under SD & HDF5 data-models | |||
| End "define" mode | ncendef | Unneccessary under SD & HDF5 data-models | |||
| Query the number of datasets, dimensions and attributes in a file | ncinquire | SDfileinfo | H5Dget_info H5Rget_num_relations H5Gget_num_contents |
HDF5 interface is more granular and flexible | |
| Update a writeable file with current changes | ncsync | AIO_flush | H5Mflush | HDF5 interface is more flexible because it can be applied to parts of the file hierarchy instead of the whole file at once. The SD interface does not have this feature, although most of the lower HDF library supports it. | |
| Close file access without applying recent changes | ncabort | How useful is this feature? | |||
| Create new dimension | ncdimdef | SDsetdimname | H5Mcreate | SD interface actually creates dimensions with datasets, this just allows naming them | |
| Get ID of existing dimension | ncdimid | SDgetdimid | H5Maccess | SD interface looks up dimensions by index and the netCDF interface uses names, but they are close enough. The HDF5 interface does not current allow access to particular dimensions, only the dataspace as a whole. | |
| Get size & name of dimension | ncdiminq | SDdiminfo | H5Mget_name H5Sget_lrank |
Only a rough match | |
| Rename dimension | ncdimrename | SDsetdimname | H5Mset_name | ||
| Create a new dataset | ncvardef | SDcreate | AIO_mkarray | H5Mcreate | |
| Attach to an existing dataset | ncvarid | SDselect | AIO_arr_load | H5Maccess | |
| Get basic information about a dataset | ncvarinq | SDgetinfo | AIO_arr_get_btype AIO_arr_get_nelmts AIO_arr_get_nbdims AIO_arr_get_bdims AIO_arr_get_slab |
H5Dget_info | All interfaces have different levels of information that they return, some use of auxilliary functions is required to get equivalent amount of information |
| Write a single value to a dataset | ncvarput1 | SDwritedata | AIO_write | H5Dwrite | What is this useful for? |
| Read a single value from a dataset | ncvarget1 | SDreaddata | AIO_read | H5Dread | What is this useful for? |
| Write a solid hyperslab of data (i.e. subset) to a dataset | ncvarput | SDwritedata | AIO_write | H5Dwrite | |
| Read a solid hyperslab of data (i.e. subset) from a dataset | ncvarget | SDreaddata | AIO_read | H5Dread | |
| Write a general hyperslab of data (i.e. possibly subsampled) to a dataset | ncvarputg | SDwritedata | AIO_write | H5Dwrite | |
| Read a general hyperslab of data (i.e. possibly subsampled) from a dataset | ncvargetg | SDreaddata | AIO_read | H5Dread | |
| Rename a dataset variable | ncvarrename | H5Mset_name | |||
| Add an attribute to a dataset | ncattput | SDsetattr | H5Rattach_oid | HDF5 requires creating a seperate object to attach to a dataset, but it also allows objects to be attributes of any other object, even nested. | |
| Get attribute information | ncattinq | SDattrinfo | H5Dget_info | HDF5 has no specific function for attributes, they are treated as all other objects in the file. | |
| Retrieve attribute for a dataset | ncattget | SDreadattr | H5Dread | HDF5 uses general dataset I/O for attributes. | |
| Copy attribute from one dataset to another | ncattcopy | What is this used for? | |||
| Get name of attribute | ncattname | SDattrinfo | H5Mget_name | ||
| Rename attribute | ncattrename | H5Mset_name | |||
| Delete attribute | ncattdel | H5Mdelete | This can be faked in current HDF interface with lower-level calls | ||
| Compute # of bytes to store a number-type | nctypelen | DFKNTsize | Hmm, the HDF5 Datatype interface needs this functionality. | ||
| Indicate that fill-values are to be written to dataset | ncsetfill | SDsetfillmode | HDF5 Datatype interface should work on this functionality | ||
| Get information about "record" variables (Those datasets which share the same unlimited dimension | ncrecinq | This should probably be wrapped in a higher layer interface, if it's needed for HDF5. | |||
| Get a record from each dataset sharing the unlimited dimension | ncrecget | This is somewhat equivalent to reading a vdata with non-interlaced fields, only in a dataset oriented way. This should also be wrapped in a higher layer interface if it's necessary for HDF5. | |||
| Put a record from each dataset sharing the unlimited dimension | ncrecput | This is somewhat equivalent to writing a vdata with non-interlaced fields, only in a dataset oriented way. This should also be wrapped in a higher layer interface if it's necessary for HDF5. | |||
| Map a dataset's name to an index to reference it with | SDnametoindex | H5Mfind_name | Equivalent functionality except HDF5 call returns an OID instead of an index. | ||
| Get the valid range of values for data in a dataset | SDgetrange | Easily implemented with attributes at a higher level for HDF5. | |||
| Release access to a dataset | SDendaccess | AIO_arr_destroy | H5Mrelease | Odd that the netCDF API doesn't have this... | |
| Set the valid range of data in a dataset | SDsetrange | Easily implemented with attributes at a higher level for HDF5. | |||
| Set the label, units, format, etc. of the data values in a dataset | SDsetdatastrs | Easily implemented with attributes at a higher level for HDF5. | |||
| Get the label, units, format, etc. of the data values in a dataset | SDgetdatastrs | Easily implemented with attributes at a higher level for HDF5. | |||
| Set the label, units, format, etc. of the dimensions in a dataset | SDsetdimstrs | Easily implemented with attributes at a higher level for HDF5. | |||
| Get the label, units, format, etc. of the dimensions in a dataset | SDgetdimstrs | Easily implemented with attributes at a higher level for HDF5. | |||
| Set the scale of the dimensions in a dataset | SDsetdimscale | Easily implemented with attributes at a higher level for HDF5. | |||
| Get the scale of the dimensions in a dataset | SDgetdimscale | Easily implemented with attributes at a higher level for HDF5. | |||
| Set the calibration parameters of the data values in a dataset | SDsetcal | Easily implemented with attributes at a higher level for HDF5. | |||
| Get the calibration parameters of the data values in a dataset | SDgetcal | Easily implemented with attributes at a higher level for HDF5. | |||
| Set the fill value for the data values in a dataset | SDsetfillvalue | HDF5 needs something like this, I'm not certain where to put it. | |||
| Get the fill value for the data values in a dataset | SDgetfillvalue | HDF5 needs something like this, I'm not certain where to put it. | |||
| Move/Set the dataset to be in an 'external' file | SDsetexternalfile | H5Dset_storage | HDF5 has simple functions for this, but needs an API for setting up the storage flow. | ||
| Move/Set the dataset to be stored using only certain bits from the dataset | SDsetnbitdataset | H5Dset_storage | HDF5 has simple functions for this, but needs an API for setting up the storage flow. | ||
| Move/Set the dataset to be stored in compressed form | SDsetcompress | H5Dset_storage | HDF5 has simple functions for this, but needs an API for setting up the storage flow. | ||
| Search for an dataset attribute with particular name | SDfindattr | H5Mfind_name H5Mwild_search |
HDF5 can handle wildcard searchs for this feature. | ||
| Map a run-time dataset handle to a persistant disk reference | SDidtoref | I'm not certain this is needed for HDF5. | |||
| Map a persistant disk reference for a dataset to an index in a group | SDreftoindex | I'm not certain this is needed for HDF5. | |||
| Determine if a dataset is a 'record' variable (i.e. it has an unlimited dimension) | SDisrecord | Easily implemented by querying the dimensionality at a higher level for HDF5. | |||
| Determine if a dataset is a 'coordinate' variable (i.e. it is used as a dimension) | SDiscoord | I'm not certain this is needed for HDF5. | |||
| Set the access type (i.e. parallel or serial) for dataset I/O | SDsetaccesstype | HDF5 has functions for reading the information about this, but needs a better API for setting up the storage flow. | |||
| Set the size of blocks used to store a dataset with unlimited dimensions | SDsetblocksize | HDF5 has functions for reading the information about this, but needs a better API for setting up the storage flow. | |||
| Sets backward compatibility of dimensions created. | SDsetdimval_comp | Unneccessary in HDF5. | |||
| Checks backward compatibility of dimensions created. | SDisdimval_comp | Unneccessary in HDF5. | |||
| Move/Set the dataset to be stored in chunked form | SDsetchunk | H5Dset_storage | HDF5 has simple functions for this, but needs an API for setting up the storage flow. | ||
| Get the chunking information for a dataset stored in chunked form | SDgetchunkinfo | H5Dstorage_detail | |||
| Read/Write chunks of a dataset using a chunk index | SDreadchunk SDwritechunk |
I'm not certain that HDF5 needs something like this. | |||
| Tune chunk caching parameters for chunked datasets | SDsetchunkcache | HDF5 needs something like this. | |||
| Change some default behavior of the library | AIO_defaults | Something like this would be useful in HDF5, to tune I/O pipelines, etc. | |||
| Flush and close all open files | AIO_exit | Something like this might be useful in HDF5, although it could be encapsulated with a higher-level function. | |||
| Target an architecture for data-type storage | AIO_target | There are some rough parallels with using the data-type in HDF5 to create data-type objects which can be used to write out future datasets. | |||
| Map a filename to a file ID | AIO_filename | H5Mget_name | |||
| Get the active directory (where new datasets are created) | AIO_getcwd | HDF5 allows multiple directories (groups) to be attached to, any of which can have new datasets created within it. | |||
| Change active directory | AIO_chdir | Since HDF5 has a slightly different access method for directories (groups), this functionality can be wrapped around calls to H5Gget_oid_by_name. | |||
| Create directory | AIO_mkdir | H5Mcreate | |||
| Return detailed information about an object | AIO_stat | H5Dget_info H5Dstorage_detail |
Perhaps more information should be provided through another function in HDF5? | ||
| Get "flag" information | AIO_getflags | Not required in HDF5. | |||
| Set "flag" information | AIO_setflags | Not required in HDF5. | |||
| Get detailed information about all objects in a directory | AIO_ls | H5Gget_content_info_mult H5Dget_info H5Dstorage_detail |
Only roughly equivalent functionality in HDF5, perhaps more should be added? | ||
| Get base type of object | AIO_BASIC | H5Gget_content_info | |||
| Set base type of dataset | AIO_arr_set_btype | H5Mcreate(DATATYPE) | |||
| Set dimensionality of dataset | AIO_arr_set_bdims | H5Mcreate(DATASPACE) | |||
| Set slab of dataset to write | AIO_arr_set_slab | This is similar to the process of creating a dataspace for use when performing I/O on an HDF5 dataset | |||
| Describe chunking of dataset to write | AIO_arr_set_chunk | H5Dset_storage | |||
| Describe array index permutation of dataset to write | AIO_arr_set_perm | H5Dset_storage | |||
| Create a new dataset with dataspace and datatype information from an existing dataset. | AIO_arr_copy | This can be mimicked in HDF5 by attaching to the datatype and dataspace of an existing dataset and using the IDs to create new datasets. | |||
| Create a new directory to group objects within | AIO_mkgroup | H5Mcreate(GROUP) | |||
| Read name of objects in directory | AIO_read_group | H5Gget_content_info_mult | |||
| Add objects to directory | AIO_write_group | H5Ginsert_item_mult | |||
| Combine an architecture and numeric type to derive the format's datatype | AIO_COMBINE | This is a nice feature to add to HDF5. | |||
| Derive an architecture from the format's datatype | AIO_ARCH | This is a nice feature to add to HDF5. | |||
| Derive a numeric type from the format's datatype | AIO_PNT | This is a nice feature to add to HDF5. | |||
| Register error handling function for library to call when errors occur | AIO_error_handler | This should be added to HDF5. |